What a customer interview can measure, and what it never will

People do not usually lie in customer interviews. They answer the question you asked. The trouble is that most questions founders ask have a polite answer built into them, and the polite answer is free to give.

So treat the question as the instrument. A bent instrument gives you a clean-looking reading of nothing.

What to take away

What an interview can actually measure

An interview is good at recovering the past and bad at predicting the future. That single limit decides almost everything about how you should run one.

You can get a reliable account of what someone did last Tuesday, what it cost them, what they tried first, what they gave up on, and who else was involved.

You cannot get a reliable account of what they will buy in March. They are not evasive; they genuinely do not know. Answering feels like being helpful, not like being asked to predict.

Three things people report badly about themselves:

The leading question, and four ways it leaks

A leading question hands over the answer and asks for a signature. It usually leaks in one of these ways.

It names the problem for them. "Do you find scheduling frustrating?" now contains the word frustrating, and agreeing is cheaper than disagreeing. Ask what happened the last time they scheduled something instead.

It describes your solution first. Once someone knows what you built, everything they say afterwards is a response to your idea rather than a description of their life. Keep the product out of the conversation until you have the story you came for.

It asks for a rating on a scale you invented. "How big a problem is this, one to ten?" produces a number with no unit. Ask what they spent on it instead: hours, money, arguments, workarounds.

It offers a hypothetical. "If there were a tool that did this, would you try it?" Every honest person says yes. Yes costs nothing.

None of this is a founder's failing. How heavily the wording and order of a question move the answer that comes back is documented at length in the guidance on writing survey questions, and the effects are large enough to survive professional question design.

Here is the shape of the repair. The right-hand column is not a script. It is the move: replace an opinion request with a request for a specific past event.

Running the conversation

Ask for one story, not a survey. "Tell me about the last time" is the workhorse. One concrete episode, followed all the way through, is worth more than twenty general impressions. When they generalize, pull them back: "Was that the case in the last one too?"

Let the silence run. Most people give you the tidy answer first, stop, and then keep talking if you do not fill the gap. The second thing they say is usually less rehearsed. Counting to four before you speak is an unglamorous technique that works.

Follow the friction, not the feature. When someone mentions a workaround, stop and stay there. A spreadsheet somebody maintains by hand, a message they send twice, a step they do outside the official system. Workarounds are the fossil record of an unsolved problem, and they cost the person something real, which is why they are trustworthy.

Ask what they already tried. Someone who has tried three things and abandoned all three is telling you the problem is worth effort. Someone who has never looked is telling you it is not, whatever they say next.

Notice who is not in the room. The person describing the problem and the person who can authorize a fix are often different people. Ask early who else would be involved, and what happened the last time that person was asked for money.

Do not defend anything. The moment you explain why they misunderstood, the interview is over and a sales call has started. Write the objection down and thank them.

Compliments are a warning sign

Enthusiasm in an interview costs the speaker nothing. It is the cheapest thing a person can give you and it feels like the most valuable, which is a bad combination.

Test enthusiasm by making it slightly expensive while still on the call. Ask for something small and real: an introduction to a colleague or a copy of the spreadsheet they described. Then request half an hour next week to look at it together, or permission to email them when there is something to try.

Not a payment. Just a cost.

The person who says the idea is brilliant and will not spend fifteen minutes on it has told you something. The person who is lukewarm but pulls up their calendar has told you something else. Believe the calendar.

Who counts as a valid interview

Two filters, applied before you book anything.

Friends and former colleagues are not useless. They are just biased in a known direction, and they are a channel you cannot scale. Use them to sharpen your questions. Do not use them to decide.

Reading what you collected

Write up each conversation within a day, while you still remember what the pauses meant. Separate three things in your notes and never let them merge:

The third one is yours. It is the part that will be wrong. Keeping it in its own column is what lets you go back later and re-read the evidence when your conclusion changes.

Then look for the same detail appearing in different lives. One person's workaround is an anecdote. The same workaround, invented separately by four people who do not know each other, is a design specification. Patterns across independent sources are the only kind of pattern worth acting on, and they show up faster than you expect once the questions stop leading.

What would change your mind

Before a batch of interviews, write down the sentence you are hoping to hear and the sentence that would make you stop. If there is no sentence anyone could say that would disappoint you, you are not running a test. You are collecting quotes.

This is the honest way to decide when to stop interviewing, not a target number. Stop when new conversations stop surprising you.

Stop when the next real question can only be answered by asking someone for money or time, not an opinion. At that point interviews have given you what they can, and the work moves to offer testing and a price someone has to refuse.

What interviews cannot settle

They cannot tell you whether people will pay. Only a price and a refusal can do that.

They cannot tell you the market size, because interviews are depth, not breadth. A few good conversations convince you of something true for six people and false for six thousand.

Market validation is a separate job with separate evidence, drawn from statistical sources, not conversations. The SBA's outline of market research and competitive analysis states plainly what that job involves.

They cannot tell you whether you can build or deliver it. That question belongs to you, and no customer can answer it.

They cannot tell you whether you asked the wrong people, and nothing inside the room can. This error survives every good interview technique, so check who you talked to before you trust what they said.

Whether the problem itself was described honestly is upstream of all of it, in problem discovery. Whether you can afford the answer is a separate matter again, in founder readiness.

Common questions

How many interviews are enough?

There is no number. Stop when several conversations in a row change nothing in your written problem statement and you can predict what the next person will say about their current method. That is saturation for the people you have been able to reach, which is not the same as saturation for the market.

Can I interview people I already know?

Yes, for sharpening questions and learning vocabulary. Keep them in a separate column, because a route that ran through your own address book teaches you about the problem and nothing about whether the audience is reachable.

What do I do when somebody asks what I am building?

Say you would rather hear about their situation first and will explain at the end. Almost nobody objects. Keep anything said after the reveal in its own section of the notes, because it is a reaction to your idea rather than an account of their life.

Is it worth interviewing people who did not buy from a competitor?

It is the most informative group available and the hardest to find. People who looked and declined can tell you what the category is failing to do, which current users have already made peace with.

What if every conversation contradicts the last one?

You are looking at several segments rather than one. Split them and write a separate problem statement for each rather than averaging into a description that fits nobody.